Multi-label Learning via Supervised Autoencoder

Multi-label learning has been applied in various areas. One crucial research issue in multi-label learning is how to find a complex nonlinear mapping between instances and multi-label such that the instances with unknown labels can be predicted with the mapping. In this paper, we propose a multi-label learning method based on supervised deep autoencoder, and study the effects of the various combinations of the constraints of hidden layers and output layers. In the output layer, the sum-square error of the true labels and estimating labels is minimized through backpropagation-through-time learning algorithm. Extensive experiments conducted on eight real-world datasets demonstrate the effectiveness of our proposed method compared with several state-of-art baseline methods.